---
license: apache-2.0
datasets:
- MidhunKanadan/logical-fallacy-classification
language:
- en
metrics:
- accuracy
- precision
- recall
- f1
base_model:
- roberta-large
pipeline_tag: text-classification
library_name: transformers
tags:
- fallacy-detection
- logical-fallacies
- text-classification
- transformers
- roberta
- fallacy-classification
---

# roberta-large-fallacy-classification

This model is a fine-tuned version of `roberta-large` trained on the [Logical Fallacy Classification Dataset](https://huggingface.co/datasets/MidhunKanadan/logical-fallacy-classification).  It is capable of classifying various types of logical fallacies in text.

## Model Details

- **Base Model**: `roberta-large`
- **Dataset**: Logical Fallacy Dataset
- **Number of Classes**: 13
- **Training Parameters**:
  - **Learning Rate**: 2e-6
  - **Batch Size**: 8 (gradient accumulation for an effective batch size of 16)
  - **Weight Decay**: 0.01
  - **Training Epochs**: 15
  - **Mixed Precision (FP16)**: Enabled
- **Features**:
  - Class weights to handle dataset imbalance
  - Tokenization with truncation and padding (maximum length: 128)

## Supported Fallacies

The model can classify the following types of logical fallacies:
1. **Equivocation**
2. **Faulty Generalization**
3. **Fallacy of Logic**
4. **Ad Populum**
5. **Circular Reasoning**
6. **False Dilemma**
7. **False Causality**
8. **Fallacy of Extension**
9. **Fallacy of Credibility**
10. **Fallacy of Relevance**
11. **Intentional**
12. **Appeal to Emotion**
13. **Ad Hominem**

## Text Classification Pipeline

To use the model for quick classification with a text pipeline:

```python
from transformers import pipeline

pipe = pipeline("text-classification", model="MidhunKanadan/roberta-large-fallacy-classification", device=0)
text = "The rooster crows always before the sun rises, therefore the crowing rooster causes the sun to rise."
result = pipe(text)[0]
print(f"Predicted Label: {result['label']}, Score: {result['score']:.4f}")
```

Expected Output:
```
Predicted Label: false causality, Score: 0.9632
```

## Advanced Usage: Predict Scores for All Labels

```python
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch.nn.functional as F

model_path = "MidhunKanadan/roberta-large-fallacy-classification"
text = "The rooster crows always before the sun rises, therefore the crowing rooster causes the sun to rise."

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForSequenceClassification.from_pretrained(model_path).to("cuda")
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=128).to("cuda")

with torch.no_grad():
    probs = F.softmax(model(**inputs).logits, dim=-1)
    results = {model.config.id2label[i]: score.item() for i, score in enumerate(probs[0])}

# Print scores for all labels
for label, score in sorted(results.items(), key=lambda x: x[1], reverse=True):
    print(f"{label}: {score:.4f}")
```

Expected Output:
```
false causality: 0.9632
fallacy of logic: 0.0139
faulty generalization: 0.0054
intentional: 0.0029
fallacy of credibility: 0.0023
equivocation: 0.0022
fallacy of extension: 0.0020
ad hominem: 0.0019
circular reasoning: 0.0016
false dilemma: 0.0015
fallacy of relevance: 0.0013
ad populum: 0.0009
appeal to emotion: 0.0009
```

## Dataset

- **Dataset Name**: Logical Fallacy Classification Dataset
- **Source**: [Logical Fallacy Classification Dataset](https://huggingface.co/datasets/MidhunKanadan/logical-fallacy-classification)
- **Number of Classes**: 13 fallacies (e.g., ad hominem, appeal to emotion, faulty generalization, etc.)

## Applications

- **Education**: Teach logical reasoning and critical thinking by identifying common fallacies.
- **Argumentation Analysis**: Evaluate the validity of arguments in debates, essays, and articles.
- **AI Assistants**: Enhance conversational AI systems with critical reasoning capabilities.
- **Content Moderation**: Identify logical flaws in online debates or social media discussions.

## License

The model is licensed under the Apache 2.0 License.